Uncertainty of trading rules in currency markets: an application of non-parametric bootstrapping

The problem of measuring the precision of signals generated by fundamental macroeconomic models is not trivial. In this paper, we suggest three different approaches for the estimation of the true and unknown distribution of the population signal. We apply the bootstrapping procedure described by Efron and Tibshirani (Stat. Sci. 1 (1986) 54) to estimate the empirical distribution of the signal and measure its precision at a specific point in time with confidence intervals. Direct and indirect bootstrapping methods are devised as a way to capture the unknown variability of the signal without altering the information content of the available data. This framework is then implemented for a simple fundamental model for the CAD/$ exchange rate. We find that accounting for skewness and prediction bias affects significantly the shape and width of the estimated confidence intervals around the estimated directional signal, and that the two proposed forms of bootstrapping are more satisfactorily than a naı̈ve Historical approach in highlighting the uncertainty surrounding the model’s predictions, and generating a measure of precision in the resulting model’s recommendations. © 2002 Elsevier Science B.V. All rights reserved. JEL classification: F31; F37